DWAS and Toenail metals from the NHBCS
Julia Bauer
11/30/2020
Research question
What is the contribution of diet to toenail metals concentrations in three-year-old participants of the New Hampshire Birth Cohort Study?
Plan
Explore relationships between foods and toenail metals using Spearman Correlation Coefficients, make correlogram
Estimate overall contribution of diet to variation in toenail metals using multiple adjusted R2
Estimate effect of [amount] increase in specific food per day on toenail metal concentration, adjusting for covariates. Use multivariable linear regression
Ln transform all biomarkers to improve model fit, normalize residualsExponentiate coefficients to represent percent change in biomarker concentration based on increase of 10 grams per day(?) for each food
Principal component analysis to group food and predict different toenail metals?
Table 1. Summary Statistics
| Characteristic | N | N = 5321 |
|---|---|---|
| Male | 532 | 274 (52%) |
| Maternal smoking status | 511 | |
| 0 | 459 (90%) | |
| 1 | 22 (4.3%) | |
| 2 | 30 (5.9%) | |
| Maternal age (years) | 532 | 31.1 (28.3, 34.4) |
| Child age at questionnaire (days) | 532 | 1,086 (1,077, 1,094) |
| Season of toenail collection | 531 | |
| 1 | 123 (23%) | |
| 2 | 148 (28%) | |
| 3 | 146 (27%) | |
| 4 | 114 (21%) | |
| Toenail sample weight (ng) | 531 | 0.010 (0.006, 0.016) |
| Al_T | 532 | 55.43 (27.39, 109.71) |
| As_T | 532 | 0.32 (0.17, 0.55) |
| Cd_T | 532 | 0.04 (0.02, 0.08) |
| Co_T | 492 | 0.05 (0.03, 0.11) |
| Cr_T | 532 | 0.53 (0.29, 0.99) |
| Cu_T | 532 | 5.94 (4.13, 8.66) |
| Fe_T | 532 | 87.10 (46.35, 190.10) |
| Hg_T | 532 | 0.02 (0.01, 0.05) |
| Mn_T | 532 | 1.44 (0.56, 3.26) |
| Mo_T | 532 | 0.02 (0.02, 0.04) |
| Ni_T | 532 | 14.07 (2.95, 57.59) |
| Pb_T | 532 | 0.89 (0.45, 1.82) |
| Sb_T | 532 | 0.07 (0.05, 0.12) |
| Se_T | 532 | 0.92 (0.84, 1.01) |
| Sn_T | 532 | 0.40 (0.23, 0.73) |
| U_T | 236 | 0.03 (0.01, 0.07) |
| V_T | 532 | 0.09 (0.04, 0.18) |
| Al_W | 523 | 10.00 (10.00, 10.00) |
| As_W | 523 | 0.04 (0.02, 0.04) |
| Cd_W | 523 | 0.02 (0.01, 0.03) |
| Co_W | 523 | 0.01 (0.01, 0.01) |
| Cr_W | 515 | 0.60 (0.10, 0.60) |
| Cu_W | 523 | 0.30 (0.03, 0.30) |
| Fe_W | 523 | 10.00 (10.00, 10.00) |
| Hg_W | 316 | 0.01 (0.01, 0.03) |
| Mn_W | 523 | 0.10 (0.10, 0.10) |
| Mo_W | 520 | 0.04 (0.04, 0.04) |
| Ni_W | 523 | 0.30 (0.03, 0.30) |
| Pb_W | 500 | 0.01 (0.01, 0.01) |
| Sb_W | 523 | 0.01 (0.01, 0.03) |
| Se_W | 515 | 0.08 (0.04, 0.08) |
| Sn_W | 506 | 0.03 (0.03, 0.05) |
| U_W | 509 | 0.01 (0.01, 0.01) |
| V_W | 523 | 0.100 (0.100, 0.100) |
|
1
Statistics presented: n (%); Median (IQR)
|
||
Questions about Table 1:
- Units for Toenail metals concentrations - ug/g?
- Units for water metals - variable names are metal_m24G_dl, is that ug/dL?
- Maternal smoking variable: smoking_status, 0=nonsmoker, 1=former, 2=current?
- Season of collection: 1=spring - 4=winter?
- Water metals- some of the metals don’t have much variation. Wondering about that - Al, Mn, Mo, Pb
Notes on univariate statistics
- From histograms, all toenail metals distributions are skewed except Se and Zn.
- Decided to ln-transform all metals data for consistency
- BMI could be an important confounder, but we are missing n>200 so I decided not to include it in regression models (below), but it might be a good sensitivity analysis to include BMI and compare results with and without it adjusted for in a subsample with complete data later down the line.
2. Frequency dietary data statistics
After reviewing summary statistics, there were several food items that kids rarely ate, so the likelihood of metals from these foods contributing to their toenail metal concentration is extremely low and I chose not to include them in regression models. See these summary tables in excel sheets (“food frequency final tables”).
Take away’s from these tables:
- More than 250 kids frequently eat (3x or more per week):
milk, yogurt, real cheese, bananas, apple/pears, crackers, peanut butter, real butter
- More than 250 kids frequently eat (3x or more per week):
Decision to exclude 22 variables from the regression analysis due to infrequent consumption. Exclude if >300 participants reported never + once/month consumption of a given item.
Excluded foods: Sweet cereal, cooked cereal, fake cheese, cream cheese, fruit mix, vegetable stew, dry beans, chilibeans, unbeansoup, other fish, fried fish, lunchables, beef soup, tacos, white bread, biscuit/muffin, tortillas, margarine, any pie, pudding, other candy, orange juice
Decided to retain fish variables for Hg models, given fish is a known dietary source for Hg.
Correlation plots
Some moderate correlations between toenail metals and dietary variables. You can use zoom, and then pan (cross arrows symbol) for a closer look. Nice to know, but I’m thinking it might be cleaner to leave it out of manuscript and focus on regression models.
- Take-aways:
- Bananas moderately correlated with: Al_T (0.3, p = 0.02), As_T (0.4 , p = 0.05), Co_T (0.3, p = 0.06), Fe_T (0.3, p = 0.02)
- Apples/pears moderately correlated with: Al_T (0.3, p = 0.1 ), As_T (0.4 , p = 0.03), Co_T (0.5, p = 0.12), Fe_T (0.4, p = 0.1), Hg_T (0.4, p = 0.09 ), Pb_T (0.3, p = 0.09)
- Carrots: As_T (0.3, p = 0.04)
- Peanut butter: Hg_T (0.3, p =0.02 )
- Spinach: Se_T (0.3, p = 0.07)
- Bananas moderately correlated with: Al_T (0.3, p = 0.02), As_T (0.4 , p = 0.05), Co_T (0.3, p = 0.06), Fe_T (0.3, p = 0.02)
3. Estimate overall contribution of diet to variation in toenail metals using multiple adjusted R2
Adjusting base model for:
sex, age at ffq, smoking status of mother, toe nail sample weight, season of toenail collection (max?), metal in waterBase model Example:
lm1<-lm(ln.toe.Al ~ bbymale + ffq_i3Y_start_age_days + smoking_status+toe_i36M_sample_wgt + toe_i36M_max_collection_season + fw_m24G_dl_Al, data=dietnew1)Then ran base + 65 diet variables for additional dietary adjustment
Table 2. Amount of variation in 3-year-olds’ toenail metals explained by diet*
| Toenail metal | Adj. r-sq base model | Adj. r-sq base + diet model | Percent of variation explained by diet |
|---|---|---|---|
| Aluminum | 0.25 | 0.33 | 0.08 |
| Arsenic | 0.07 | 0.16 | 0.09 |
| Cadmium | 0.04 | 0.19 | 0.15 |
| Chromium | 0.12 | 0.25 | 0.13 |
| Copper | 0.03 | 0.16 | 0.13 |
| Iron | 0.25 | 0.33 | 0.07 |
| Mercury | 0.05 | 0.27 | 0.22 |
| Manganese | 0.15 | 0.23 | 0.08 |
| Molybdenum | 0.09 | 0.20 | 0.11 |
| Nickel | 0.02 | 0.15 | 0.13 |
| Lead | 0.06 | 0.22 | 0.16 |
| Antimony | 0.03 | 0.17 | 0.14 |
| Selenium | 0.03 | 0.15 | 0.12 |
| Tin | 0.03 | 0.14 | 0.13 |
| Zinc | 0.01 | 0.14 | 0.13 |
| Vanadium | 0.18 | 0.28 | 0.10 |
*models adjusted for child sex, child age, season of toenail collection, toenail sample weight, smoking in the home, home tap water respective metal concentration
4. Regression Models for estimating the effect of 10 g increase in specific food per day on toenail metal concentration, adjusting for covariates.
- Exponentiated coefficients to represent percent change in biomarker concentration based on increase of 10 grams per month or per day? for each food
- models adjusted for child sex, child age, season of toenail collection, toenail sample weight, smoking in the home, home tap water respective metal concentration. Toenail metals Ln-transformed in linear models. Percent change in toenail metal concentration per 10 grams increase in each food calculated as: (exp(beta)-1) x 100 x 10g.
Food presented here were selected based on size of beta and p-value value, respectively.
Values from figure
| Metal | Food | Estimated percent change | 95%CI | |
|---|---|---|---|---|
| Al | Cold cereal | 14.92 | 1.14 | 28.89 |
| Al | Hot dogs | 14.24 | -3.78 | 32.60 |
| Al | Popsicles | 6.25 | 0.83 | 11.69 |
| As | Bacon/sausage | 24.00 | -3.89 | 52.68 |
| As | Cold cereal | 13.83 | 0.10 | 27.75 |
| As | Crackers | 13.73 | -16.79 | 45.20 |
| Cd | Tomatoes | 32.98 | 6.50 | 60.15 |
| Cd | Cold cereal | 20.59 | 3.39 | 38.09 |
| Cd | Beef | 19.64 | -9.09 | 49.21 |
| Hg | Fried fish | 62.96 | 9.71 | 119.01 |
| Hg | Salty snacks | 31.24 | -10.93 | 75.20 |
| Hg | Pork | 28.57 | -18.94 | 78.38 |
| Ni | Burgers | 60.36 | -11.94 | 137.94 |
| Ni | Pretzels | 52.47 | -21.69 | 132.26 |
| Ni | Rolls/muffins | 45.80 | 9.41 | 83.51 |
| Pb | Tomatoes | 22.60 | -2.47 | 48.30 |
| Pb | Peanut butter | 23.95 | -2.34 | 50.94 |
| Pb | Pork | 22.66 | -10.79 | 57.24 |
| Sb | Green salad | 13.93 | -14.67 | 43.36 |
| Sb | Tortillas | 19.08 | -14.31 | 53.60 |
| Sb | Bacon/sausage | 16.32 | -7.85 | 41.08 |
| Sn | Bacon/sausage | 17.64 | -14.44 | 50.76 |
| Sn | Raisin/dried fruit | 13.84 | -16.60 | 45.22 |
| Sn | Beef | 13.67 | -11.72 | 39.70 |
| Cr | Bacon/sausage | 21.60 | -4.60 | 48.40 |
| Cr | Pancakes/waffles | 11.00 | -0.70 | 22.90 |
| Cr | Broccoli | 8.50 | -2.10 | 19.30 |
| Cu | Cheese | 8.90 | -0.33 | 18.23 |
| Cu | Green beans | 10.70 | 1.20 | 20.10 |
| Cu | Tomatoes | 9.80 | -3.00 | 22.70 |
| Fe | Cold cereal | 13.40 | 0.30 | 26.60 |
| Fe | Hot dogs | 20.10 | 2.90 | 37.70 |
| Fe | Popsicles | 5.90 | 0.70 | 11.00 |
| Mn | Cold cereal | 18.10 | 0.30 | 36.30 |
| Mn | Green beans | 20.10 | -0.10 | 40.80 |
| Mn | Hot dogs | 20.00 | -3.50 | 44.00 |
| Mo | Burgers | 27.33 | 0.40 | 54.90 |
| Mo | Pork | 14.30 | -8.00 | 37.10 |
| Mo | Bacon/sausage | 12.20 | -10.80 | 35.80 |
| Se | Jelly/honey/sugar | 8.50 | 1.00 | 16.00 |
| Se | Nuts/seeds | 4.30 | -1.20 | 9.80 |
| Se | Cold cereal | 2.10 | 0.40 | 4.60 |
| Zn | Salty snacks | 6.20 | -1.80 | 14.30 |
| Zn | Nuts/seeds | 5.70 | -4.90 | 16.40 |
| Zn | Broccoli | 3.90 | 0.10 | 7.80 |